Predicting surgical intensive care unit readmission with machine learning model: Bi-center training and validation.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Critical Care Medicine

Background:

• Surgical intensive care unit (SICU) readmissions are associated with high mortality and costs.
• Predicting SICU readmission risk is critical for proactive patient management.
• Developing accurate predictive models can improve patient outcomes and resource allocation.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for predicting SICU readmission.
• To compare the performance of various ML algorithms against traditional methods.
• To identify key clinical factors associated with SICU readmission.

Main Methods:

• Retrospective analysis of electronic healthcare records from two hospital branches.
• Development and internal validation of ML models including Logistic Regression, Random Forest, Gradient Boosting, Artificial Neural Networks, and Support Vector Machines.
• External validation of the best performing model against traditional logistic regression methods.

Main Results:

• The Gradient Boosting (GB) model demonstrated superior performance with an AUROC of 0.82 in internal validation.
• Key predictors of SICU readmission included central venous catheter usage, pre-ICU stay duration, blood urea nitrogen, and carbapenem usage.
• The GB model outperformed traditional logistic regression methods in the external validation cohort.

Conclusions:

• Machine learning models, particularly Gradient Boosting, offer enhanced accuracy and reliability for predicting SICU readmission.
• ML models provide a valuable tool for identifying high-risk patients, enabling targeted interventions.
• The findings support the integration of ML into clinical practice for critical care patient management.